University of Central Florida University of Central Florida
STARS
STARS
Electronic Theses and Dissertations, 2004-2019
2014
Analysis of taxi drivers' driving behavior based on a driving
Analysis of taxi drivers' driving behavior based on a driving
simulator experiment
simulator experiment
Jiawei WuUniversity of Central Florida
Part of the Transportation Engineering Commons Find similar works at: https://stars.library.ucf.edu/etd
University of Central Florida Libraries http://library.ucf.edu
This Masters Thesis (Open Access) is brought to you for free and open access by STARS. It has been accepted for inclusion in Electronic Theses and Dissertations, 2004-2019 by an authorized administrator of STARS. For more information, please contact [email protected].
STARS Citation STARS Citation
Wu, Jiawei, "Analysis of taxi drivers' driving behavior based on a driving simulator experiment" (2014). Electronic Theses and Dissertations, 2004-2019. 4587.
ANALYSIS OF TAXI DRIVERS’DRIVING BEHAVIOR
BASED ON A DRIVING SIMULATOR EXPERIMENT
by
JIAWEI WU
B.S. Beijing Jiaotong University, June 2012
A dissertation submitted in partial fulfillment of requirements
for the degree of Master of Science
in the Department of Civil and Environmental Engineering
in the College of Engineering and Computer Science
at the University of Central Florida
Orlando, Florida
Fall Term
2014
© 2014 Jiawei Wu
ABSTRACT
Due to comfort, convenience, and flexibility, taxis become more and more prevalent in China, especially in large cities. According to a survey reported by Beijing Traffic Development Research Center, there were 696 million taxi person-rides in Beijing in 2011. However, many violations and road crashes that were related to taxi drivers occurred more frequently. The survey showed that there were a total of 17,242 taxi violations happened in Beijing in only one month in 2003, which accounted for 56% of all drivers’ violations. Besides, taxi drivers also had a larger accident rate than other drivers, which showed that nearly 20% of taxi drivers had accidents each year. This study mainly focuses on investigating differences in driving behavior between taxi drivers and non-professional drivers.
To examine the overall characteristics of taxi drivers and non-professional drivers, this study applied a hierarchical driving behavior assessment method to evaluate driving behaviors. This method is divided into three levels, including low-risk level, medium-risk level, and high-risk level. Low-risk level means the basic vehicle control. Medium-risk level refers to the vehicle dynamic decision. High-risk level represents the driver avoidance behavior when facing a potential crash.
The Beijing Jiatong University (BJTU) driving simulator was applied to test different risk level scenarios which purpose is to find out the differences between taxi drivers and non-professional drivers on driving behaviors. Nearly 60 subjects, which include taxi drivers and non-professional drivers, were recruited in this experiment. Some statistical methods were applied to analyze the data and a logistic regression model was used to perform the high-risk level.
The results showed that taxi drivers have more driving experience and their driving style is more conservative in the basic vehicle control level. For the car following behavior, taxi drivers have smaller following speed and larger gap compared to other drivers. For the yellow indication judgment behavior, although taxi drivers are slower than non-professional drivers when getting into the intersection, taxi drivers are more likely to run red light. For the lane changing behavior, taxi drivers’ lane changing time is longer than others and lane changing average speed of taxi drivers is lower than other drivers.
Another different behavior in high-risk level is that taxi drivers are more inclined to turn the steering wheel when facing a potential crash compared to professional drivers. However, non-professional drivers have more abrupt deceleration behaviors if they have the same situation.
According to the experiment results, taxi drivers have a smaller crash rate compared to non-professional drivers. Taxi drivers spend a large amount of time on the road so that their driving experience must exceed that of non-professional drivers, which may bring them more skills. It is also speculated that because taxi drivers spend long hours on the job they probably have developed a more relaxed attitude about congestion and they are less likely to be candidates for road rage and over aggressive driving habits.
To my parents
Ruiming Wu and Jingli Bai
Who encourage me to realize my dream
ACKNOWLEDGMENT
I would first sincerely express my gratitude to my advisor Dr. Essam Radwan and Dr. Xuedong Yan. Their support and guidance contributed greatly to the success of this work. What I learned from them during my master period is not only the special knowledge about transportation system, but also spirits of seriousness, honesty, modesty, and confidence as a dependent researcher.
I would like to thank my committee members, Dr. Essam Radwan, Dr. Mohammed Abdel-Aty, and Dr. Hatem Abou-Senna for serving in my committee and for providing me with excellent course instructions and guidance.
Appreciation is due my friend Bin Wang, Wang Xiang, Zhuo Yang, and Yongcun Zhao for supports and assistances with driving simulator in Beijing Jiaotong University.
Finally, thanks to my girlfriend for her support to my whole graduate study. Without her encouragement, I cannot image the difficulty in the transfer from Beijing Jiaotong University to University of Central Florida.
TABLE OF CONTENT
LIST OF FIGURES ... xi
LIST OF TABLES ... xii
CHAPTER ONE: INTRODUCTION ... 1
1.1 Background ... 1
1.2 Research Approaches ... 3
1.3 Research Objectives ... 4
1.4 Thesis Organization ... 5
CHAPTER TWO: LITERATURE REVIEW ... 6
2.1 Traffic Safety Research Related to Taxi Drivers ... 6
2.2 Driver Behavior Research Based on Driving Simulator ... 8
CHAPTER THREE: EXPERIMENT METHODOLOGY ... 12
3.1 Equipment ... 12
3.2 Experimental Design ... 13
3.2.1 Low-Risk Scenario Design ... 15
3.2.2 Medium-Risk Scenario Design ... 16
3.2.3 High-Risk Scenario Design... 22
3.3 Experiment Procedure ... 24
3.4 Subjects ... 25 vii
CHAPTER FOUR: DATA COLLECTION ... 27
4.1 Simulator Data Collection Procedure ... 27
4.2 Low-Risk Scenario Data Collection ... 27
4.2.1 Uphill and Downhill Scenario Data Collection ... 27
4.2.2 Right Angle Turn Scenario Data Collection ... 29
4.2.3 S-type Continuous Curve Scenario Data Collection ... 30
4.3 Medium-Risk Scenario Data Collection ... 30
4.3.1 Yellow Indication Judgment Scenario Data Collection ... 30
4.3.2 Car Following Scenario Data Collection ... 31
4.3.3 Lane-Changing Scenario Data Collection ... 33
4.4 High-Risk Scenario Data Collection ... 34
CHAPTER FIVE: ANALYSES OF LOW-RISK SCENARIOS ... 36
5.1 Uphill and Downhill Analysis ... 36
5.2 Right Angle Turn Analysis ... 41
5.3 S-type Continuous Curve Analysis ... 43
CHAPTER SIX: ANALYSES OF MEDIUM-RISK SCENARIOS ... 46
6.1 Yellow Indication Judgment Analysis ... 46
6.1.1 Speed at the Onset of Yellow Phase Analysis ... 46
6.1.2 RLR Analysis ... 48
6.2 Car Following Analysis ... 50
6.2.1 Constant period ... 50
6.2.2 Deceleration period ... 52
6.2.3 Acceleration period ... 54
6.3 Lane-Changing Analysis ... 57
CHAPTER SEVEN: ANALYSES OF HIGH-RISK SCENARIOS ... 60
7.1 Operating Speed Analysis ... 60
7.2 Lane Deviation Analysis ... 62
7.3 Deceleration Process Analysis ... 64
7.4 Crash Probability Analysis Based on Logistic Regression ... 66
CHAPTER EIGHT: CONCLUSIONS ... 69
8.1 Characteristics of Taxi Drivers on Low-Risk Level ... 69
8.2 Characteristics of Taxi Drivers on Medium-Risk Level ... 70
8.3 Characteristics of Taxi Drivers on High-Risk Level ... 71
8.4 Summary ... 72
APPENDIX A: LOW-RISK SCENARIO DATA COLLECTION ... 74
APPENDIX B: MEDIUM-RISK SCENARIO YELLOW INDICATION JUDGMENT SCENARIO DATA COLLECTION ... 77
APPENDIX C: MEDIUM-RISK SCENARIO CAR FOLLOWING SCENARIO DATA COLLECTION ... 82
APPENDIX D: MEDIUM-RISK SCENARIO LANE-CHANGING SCENARIO DATA
COLLECTION ... 86 APPENDIX E: HIGH-RISK SCENARIO DATA COLLECTION... 91 LIST OF REFERENCES ... 94
LIST OF FIGURES
Figure 1: BJTU driving simulator ... 13
Figure 2: Hierarchical driving behavior assessment method ... 14
Figure 3: Low-risk scenario design ... 16
Figure 4: Yellow indication judgment design ... 18
Figure 5: Car following scenario design ... 20
Figure 6: Lane-changing scenario design ... 21
Figure 7: VMS-I display ... 22
Figure 8: VMS-II display ... 22
Figure 9: Collision avoidance scenario design ... 23
Figure 10: Mean of the uphill starting speed (USS) ... 38
Figure 11: Mean of the uphill ending speed (UES) ... 39
Figure 12: Mean of uphill average speed (UAS) ... 39
Figure 13: MINITAB output: z-test for out-of-lane ratios ... 43
Figure 14: Mean of the S-type continuous curve starting speed (SCSS) ... 45
Figure 15: Mean of the speed at the onset of the yellow phase ... 48
Figure 16: Mean of the lane changing time ... 59
Figure 17: Mean of the lane changing average speed ... 59
Figure 18: The trends of the speed for taxi drivers and non-professional drivers ... 61
Figure 19: The trends of the deviation for taxi drivers and non-professional drivers ... 63
Figure 20: The influence of taxi drivers and non-professional drivers on DEC ... 66
LIST OF TABLES
Table 1: Gender and occupation distribution of each scenario ... 25 Table 2: Descriptive statistics of five factors related to the uphill period ... 37 Table 3: Analysis of variance (ANOVA) for dependent variables related to the uphill period ... 37 Table 4: Descriptive statistics of five factors related to the downhill period ... 40 Table 5: Analysis of variance (ANOVA) for dependent variables related to the downhill period ... 40 Table 6: Descriptive statistics of five factors related to the right angle turn period ... 41 Table 7: Analysis of variance (ANOVA) for dependent variables related to the right angle turn period ... 41 Table 8: Descriptive statistical results of SCSS, SCES, SCAS, and SCDE for taxi drivers and non-professional drivers... 44 Table 9: Analysis of variance (ANOVA) for dependent variables related to the S-type continuous curve period ... 45 Table 10: Descriptive statistical results of speed at the onset of yellow phase for taxi drivers and non-professional drivers... 47 Table 11: Descriptive statistical results of RLR for taxi drivers and non-professional Drivers ... 49 Table 12: Descriptive statistical results of average speed and average headway for taxi drivers and non-professional drivers ... 51 Table 13: Analysis of variance (ANOVA) of average speed and average space headway during the constant period ... 52 Table 14: Descriptive statistical results of reaction time and average deceleration for taxi drivers and non-professional drivers ... 53
Table 15: Analysis of variance (ANOVA) of reaction time and average deceleration during the deceleration period ... 54 Table 16: Descriptive statistical results of reaction time and average acceleration for taxi drivers and non-professional drivers ... 55 Table 17: Analysis of variance (ANOVA) of reaction time and average acceleration during the acceleration period ... 56 Table 18: Descriptive statistical results of lane-changing analysis for taxi drivers and
non-professional drivers ... 57 Table 19: Analysis of variance (ANOVA) of lane-changing analysis ... 58 Table 20: Descriptive statistical results for deviation at each second before the conflict time .... 61 Table 21: Analysis of variance (ANOVA) of operating speed analysis ... 62 Table 22: Descriptive statistical results for deviation at each second before the conflict time .... 63 Table 23: Analysis of variance (ANOVA) of deviation analysis ... 64 Table 24: Descriptive Statistical Results of four factors for taxi drivers and non-professional drivers ... 65 Table 25: Analysis of variance (ANOVA) table for four factors ... 65 Table 26: Logistic regression results between factors and the occurrence of a crash ... 67
CHAPTER ONE: INTRODUCTION
1.1 Background
With the rapid development of urbanization and motorization, the taxi transport industry has also
highly developed and plays an important role in the urban public transportation system in China,
especially in big cities. According to the statistics from Beijing Municipal Bureau (2008), there
are more than 69,000 taxis in Beijing, which is six times compared to 20 years ago. Meanwhile,
with the improvement of people’s living standard, taxi is getting more and more popular among
general public transportation for its comfort, convenience, and flexibility and number of those who
choose taxi for their travelling increases steadily. As reported by Beijing Traffic Development
Research Center (2012), there were 650 million taxi person-rides in Beijing in 2005, increasing to
696 million in 2011.
Generally, there is no difference between taxis and ordinary cars because of their same outlook
and vehicle performance. When collecting traffic volume data, a taxi is also considered as one
passenger-car that is the same as the ordinary car. Therefore, few studies have been conducted in
examining taxi drivers’ behavior and work conditions. In fact, in terms of driving behaviors, taxi
drivers are quite different from others. Compared to non-professional drivers, taxi drivers need to
stop, accelerate, or decelerate more frequently in order to seek customers, which may result in
some traffic problems and safety issues. The taxi industry is highly dependent on taxi drivers, who
not only safely control the vehicle but also need to provide service. According to a violation
investigation, there were a total of 17,242 taxi violations happened in Beijing in only one month
in 2003, which accounted for 56% of all drivers’ violations. The same study also indicated that
27% of taxi drivers got tickets in only one month (Chinese People’s Political Consultative
Conference (CPPCC) Member Proposal, 2006). It is speculated that the high violation rates can be
attributed to two reasons. The first reason is that the agency that grants licenses to taxi drivers is
not demanding of high standards of drivers’ qualifications. They only need to acquire normal
driver licenses and complete a short-term training before being a qualified taxi driver. The second
reason is that taxi companies employ taxi drivers and they need to pay a fairly large amount of
money each month to their companies. However, the income of taxi drivers is relatively low
compared to their workload (Shi et al., 2014). Although taxi violation phenomenon has become
slightly less in recent years, it is still a big issue.
Furthermore, taxi drivers also had a larger accident rate than non-professional drivers. The data
showed that nearly 20% of taxi drivers had accidents each year, which was a large proportion of
all traffic accidents (Shi et al., 2014). Due to the particularity of taxi industry, taxi drivers have 2
flexible work schedule, which means they can arrange their time according to their will. On the
surface, taxi drivers could have a rest when they feel tired, which may help to reduce traffic
accidents and avoid drowsy driving. However, reality is that in China taxi drivers usually work on
the average 11 hours per day and 27.8 days per month, which is far more than the legal limit of 8
hours per day and 21.75 days per week. This high work hour rate may be a reflection of the
typically taxi drivers’ work condition in China (Deng and Ou, 2009). Therefore, it can be inferred
that nearly 25% of work time for taxi drivers is drowsy driving conditions in Beijing. Although
taxi drivers often claimed and portrayed that they were aware of traffic safety requirements, they
were driving in a fatigued condition (R.Dalziel & Job, 1997), and it in many studies it was proved
to be more dangerous on the road (Connor et al., 2011; Lucidi et al., 2013; T.McCartt et al., 1996).
1.2 Research Approaches
Firstly, a literature review of relevant domain information was conducted, including traffic safety
research related to taxi drivers, and driver behavior research based on a driving simulator
experiment.
Secondly, a series of scenarios were designed in the driving simulator to collect data on both taxi
drivers’ behaviors and non-professional drivers’ behaviors. A hierarchical driving performance
assessment method was adopted during the simulator experiment, which was classified into three
different risk levels. About 55 participants were recruited in this experiment and data was collected
during the experiment.
Finally, several software packages including Microsoft EXCEL, SPSS, Minitab, and SAS were
used to analyze the data and build statistical models.
1.3 Research Objectives
The main objectives of this research are:
(1) Set up several scenarios in the driving simulator to test the drivers’ behavior, including taxi
drivers and non-professional drivers. The scenarios are divided into different levels of the
traffic safety, including the low risk (basic vehicle control), the medium risk (dynamic
decision) and the high risk (collision avoidance response behavior).
(2) Extract the drivers’ behavior from the raw data in different scenarios, such as vehicle
velocity, deceleration, reaction time and other relevant information.
(3) Find the difference between taxi drivers and non-professional drivers in driving behaviors
under different levels of traffic safety.
(4) Summarize the basic characteristics of the taxi drivers’ behavior and analyze some typical
scenarios for the traffic safety.
1.4 Thesis Organization
This chapter presents an introduction to the subject matter to be discussed as well as a description
of the research approaches and objectives. Chapter 2 delves into literature to discuss the framing
of the problem addressed by this research. Chapter 3 describes the driving simulator study
methodology, including equipment, experimental design, experiment procedure, and subject.
Chapter 4 presents the data collection procedure and describes every variable in each scenario.
Chapter 5, 6, and 7 analyze each scenario by using driving simulator data and discuss the statistical
models developed. Finally, Chapter 8 provides the conclusions and recommendations derived from
the driving simulator experiment.
CHAPTER TWO: LITERATURE REVIEW
2.1 Traffic Safety Research Related to Taxi Drivers
In recent years, many studies focused on taxi drivers’ behaviors related to the traffic safety. There
were two main aspects, including car crashes or violations. Maag et al. (1997) developed a
regression model to estimate taxi road traffic crashes for each factor by using 6 years crash data in
Canada. They found that taxi drivers have an average of 0.252 crashes per year compared to 0.07
of all drivers, which was similar to the findings of Nordjaern et al. (2012). Similarly, taxi drivers
also had a high average number of victims per crash than others. Moreover, some factors associated
with vehicle crashes and injuries were also identified among taxi drivers, such as binocular vision
problem (Maag et al., 1997), fatigue (R.Dalziel and Job, 1997), emotional well-being (Machin and
De Souza, 2004), and driving at night and driving without any passengers (Lam, 2004).
In addition, Goudine (1997) investigated the taxi drivers’ attitudes to traffic laws and penalties in
North Africa. It was found that there was a poor communication between law enforcers and taxi
drivers, which may lead to high-risk driving behaviors among taxi drivers. Another study also
attempted to identify the attitude towards traffic violation in male taxi drivers in Israeli
(Rosenbloom and Shahar, 2007). The results indicated that non-professional drivers paid more
attention to the traffic violations or penalties than taxi drivers did, especially in minor-severity
traffic regulations, which would also raise driving risks among taxi drivers. Furthermore, those
taxi drivers, who enjoyed taking more risks while driving, were more likely to commit speeding
violations (Burns and Wilde, 1995).
Some studies in developing countries also focused on the taxi drivers. In Vietnam, La et al. (2013)
used a structural questionnaire to interview drivers from five different companies during
2006-2009. They found that education on traffic rules and seat-belt usage could improve the taxi drivers’
safety and element of time pressure could also impact on crashes for taxi drivers. In Sri Lanka,
Akalanka et al. (2012) used bivariate and multivariate logistic regression to examine the
sociodemographic factors related with aggressive driving behaviors among 3-wheeler taxi drivers.
They demonstrated that the level of education and aggressive driving practice were the significant
factors that affected road traffic crashes. Besides, marital status was also associated with some
violations, such as running red lights, and drinking while driving.
Although few previous studies analyzed the effect of taxi drivers on traffic safety, most of them
only focused on the effects of taxi drivers on crashes or violations using questionnaires or crash
reports but did not pay attention to the taxi drivers’ driving behaviors. Several previous studies 7
demonstrated the potential of using driving simulators to assess traffic safety. However, none of
these studies focused on taxi drivers.
2.2 Driver Behavior Research Based on Driving Simulator
The driving simulator is a research and development experiment device used to test driving
behaviors, safety performance and the condition of the road. In general, the simulator consists of
several subsystems: a real-time vehicle simulation system, vehicle motion, visual and audio
systems, a control loading system, an operator console, and data collection system (Lee et al.,
1998). The use of an advanced driving simulator has many advantages on testing drivers’ behaviors,
including experimental control, efficiency, expense, safety, and ease of data collection (Godley et
al., 2001).
In recent years, more and more researchers used a driving simulator to test the driving behaviors
that are not easy tested in the reality. Several conditions were usually to be considered to use a
driving simulator to collect data, including environmental factors, dangerous scenarios, the new
application installed on road, and some particular groups.
Some environmental factors are difficult to test in reality, such as the adverse weather and roadside
vegetation. Therefore, a number of previous studies have focused on driving behaviors related to
the environmental factors based on driving simulator. Yan et al. (2014) investigated the effects of
foggy conditions under different risk levels on drivers’ speed behaviors according to the driving
simulator experiment. They found that drivers intended to reduce their speed under foggy
condition, but speed compensation cannot sufficiently reduce the risk of crash involvement, which
concurred with the study by Hoogendoorn et al. (2010). Broughton et al. (2007) also used a driving
simulator to test the car following behaviors under foggy conditions. They also found that drivers
were more likely to fail to maintain following distance under foggy conditions. Another study
related to the foggy condition based on driving simulator found that experienced drivers had a
larger speed than novice drivers just under clear conditions (Mueller and Trick, 2012). Under
foggy conditions, experienced drivers reduced their speed more than novice drivers. In addition,
roadside vegetation is also one of the environmental factors that are usually used driving simulator
to examine the influence on driver behavior. Fitzpatrick et al. (2014) found that trees were the
most harmful roadside objects that might lead to some crashes and drivers drove closer to the edge
line as the clear zone size increased.
Car accidents are one of the biggest issues in the transportation field. However, it is also hard to
test drivers’ response before the crash. Therefore, the driving simulator can be a good tool to assess
behavior of drivers in such a dangerous traffic situation. Several studies tested driving behaviors
for different pre-accident situations in the driving simulator (Guzek et al., 2009; Lee et al., 2002;
McGehee et al., 2000). Besides, the other dangerous situation that is suitable for testing in the
driving simulator, not in the reality is that drivers are driving while using cell phones. Alm and
Nilsson (1995) used the driving simulator to test whether there was a negative effect on drivers’
behaviors when drivers used cell phones while driving. It is found that drivers had a longer reaction
time when using cell phones and accident risk could increase because of the use of cell phones.
The similar researches related to the cell phones based on driving simulator experiment were also
studied by Strayer and Drews (2004), Lesch and Hancock (2004), Haigney et al. (2000), and Stein
et al. (1987). Similarly, there are also other dangerous situations that need to test in the driving
simulator, including work zones (Bella, 2005; McAvoy et al., 2007), and the curve condition
(Reymond et al., 2001).
Some new applications before the application in the reality also need to be tested in the driving
simulator to assess the effect of the new applications. The pavement marking, which is one of the
new applications in Florida, is to place a marking with a word “Signal Ahead” upstream of a 10
signalized intersection. This method is also tested in the driving simulator by Yan et al. (2009).
They found that pavement marking had a positive effect on drivers’ behaviors at signalized
intersection, which can reduce the probabilities of both conservative-stop and risky-go decisions.
Another example for new applications is the in-vehicle warning message to help drivers lower the
red light running violations. Some researchers also used a driving simulator to verify the
effectiveness of this new application (Philippus et al., 2009; Yan et al., 2014).
The driving simulator is also a useful tool to study the particular group on driving behaviors. For
example, drivers with Alzheimer Disease are the uncertain group for driving safety. Rizzo et al.
(1997) used high-fidelity driving simulator to measure relevant performance factors and found that
drivers with Alzheimer Disease were more than twice as likely to have crashes compared to the
normal drivers, which is the same finding with Cox et al. (1998). Besides, some other groups are
also tested in the driving simulator, including novice drivers and experienced drivers (Chan et al.,
2010; Parmet et al., 2014), old drivers and young drivers (Lee et al., 2003; Pradhan et al., 2005).
CHAPTER THREE: EXPERIMENT METHODOLOGY
3.1 Equipment
A high-fidelity driving simulator, which consists of a full-size cabin, a digital sound simulation
system, a vibration system and the center console, located in MOE Key Laboratory for Urban
Transportation Complex Systems of the Beijing Jiaotong University was used in the study, as
shown in Figure 1. It is suitable for conducting interactive driving simulation experiments under
lab-control conditions and is good for analyzing driving behavior.
The driving simulator has a linear motion base capable of operation with one degree of freedom.
It is composed of a visual system with 300 degree of front view and three rear view mirrors, a
full-size cabin of Ford Focus with real operation interface, a digital sound simulation system, a
vibration system, and the center console. The visual system allows resolution equal to 1400 x 1050
pixels for each channel. The software, including Simvista and Simcreator, is provided for modeling
road networks and driving scenarios. The data sampling frequency is up to 60 Hz.
In addition, five cameras are installed inside and outside the cabin to supervise the experimental
process. An emergency stop button is installed both inside cabin beside the driver seat and in the
front of control desk in order that either subject or researcher can discontinue the experiment
immediately in case the subject suffers driving simulation sickness.
Figure 1: BJTU driving simulator
3.2 Experimental Design
Although a number of previous studies focused on the driving behavior by using driving simulators,
most of them lack of a systematic analysis according to risk level. Yan et al. (2014) adopted a
hierarchical driving performance assessment method to evaluate driving behavior under foggy
conditions. However, they only paid attention to different speed-related situations. In this study,
an updated concept of hierarchical driving behavior assessment method was proposed to estimate
the taxi driver behavior, as shown in Figure 2. It is classified into three different risk levels. The
first level is low-risk level, which means drivers require basic vehicle control according to the
roadway alignment, such as acceleration, braking, steering. The second level is medium-risk level,
which relates to moderate risk and dynamic decision. For example, in the yellow light judgment
scenarios, drivers need to make decisions to stop or go according to the distance to the stop line
and vehicle speed. If drivers make wrong decisions like running the red light, a serious crash may
occur. Some other typical scenarios such as car following scenarios, and lane-changing scenarios
are also considered as medium-risk scenarios. The third level is high-risk level, which corresponds
to the Collision Avoidance Scenario, indicating that when an immediate crash is present, drivers
need to take emergency response to avoid crash. Since it is hard to get driver emergency reaction
data in real world, driving simulator is a good tool to evaluate driver emergency reaction in this
kind of scenario.
Figure 2: Hierarchical driving behavior assessment method
3.2.1 Low-Risk Scenario Design
To investigate basic vehicle control behaviors, the low-risk scenario was designed, as shown in
Figure 3.The low-risk scenario consisted of three sub-scenarios, including uphill and downhill
scenario (15°slope), right angle turn scenario, and S-type continuous curve scenario. All the roads
were two-way with two lanes. The speed limits were 50 km/h on the uphill and downhill segments,
and 30 km/h on the S-type continuous scenario. The subject needed to drive from the Start point
to the End point, and there was no other vehicle in front of the simulator vehicle in order that
drivers could control vehicles by themselves. Additionally, some other vehicles on the opposite
direction were added into the scenario in order to improve the scene reality.
Figure 3: Low-risk scenario design
3.2.2 Medium-Risk Scenario Design
The medium-risk scenario was divided into three parts, including the yellow indication judgment
scenario, the car following scenario, and the lane-changing scenario. Each scenario was used to
explore drivers’ behaviors related to moderate risk or dynamic decision.
3.2.2.1 Yellow Indication Judgment Scenario
To investigate driving behaviors when facing the potential red light running violation during the
yellow phase at a signalized intersection, the yellow indication judgment scenario was designed,
as shown in Figure 4, which was composed of two-lane road segments with the 80 km/h speed 16
limit. The duration of yellow signal was set to be 4.5s, which was calculated by the formula
recommended by Institute of Transportation Engineers (ITE). For the simulator experiment, if
subjects were located 5, 5.5, and 6 seconds away from the intersection stop line respectively, and
if they kept their current posted speed along their presumed path, the signal would have changed
into the yellow indication and lasted 4.5 seconds, which meant that if subjects didn’t take any
action and still drove at their present speed, they would have run the red light. Therefore, a
Time-To-Collision sensor was used to realize this scenario. TTC is defined as the time that remains until
a collision between two vehicles would have occurred if collision course and speed difference are
maintained (Minderhoud and Bovy, 2001), which was widely used in many simulation studies
(Gelau et al., 2011; Yan et al., 2014; Yan et al., 2014). TTC was also seen as the time to the
intersection stop line (TTI) in this scenario and was equal to 5, 5.5, and 6 seconds in each
intersection, respectively. The TTC sensor would trigger the traffic signal to change into the yellow
phase and last 4.5 seconds in advance of the intersection. Meanwhile, there was no other vehicle
in front of the simulator vehicle in order that drivers could decide whether to stop or go by
themselves. In order to counterbalance the temporal order effect during the experiment operation,
three intersections were randomly assigned to the road network and some other intersections,
which displayed continuous green phases, were intermingled between each test intersection, which
has the same design method as Yan et al. (2009). 17
Figure 4: Yellow indication judgment design
3.2.2.2 Car Following Scenario
The car following scenario was designed to test drivers’ car following behaviors between taxi
drivers and non-professional drivers, as shown in Figure 5. The test road was arranged on a
two-way segment with two lanes, and its total length was about 2.5 km. A platoon of system vehicles
stopped at the beginning of the scenario, waiting for the simulator vehicle to join the queue. All of
system vehicles are general vehicles, not taxis. When the simulator vehicle arrived at this point,
Sensor 1 was triggered and system vehicles would start to accelerate at 1 m/s2 along the segment.
In order to ensure the simulator vehicle could follow the leading vehicle all the time and not pass
the front vehicle in the opposite lane, several vehicles were added in the opposite direction. During
the car following process, 10 different periods, including three accelerate periods, two decelerate
periods, and five constant speed periods, were used to investigate car following behaviors in
different car following situations. Five sensors were applied to realize the scenario. Ten periods
were described as follows:
• Period 1: The leading vehicle would accelerate at 1 m/s2 until the speed was up to 40 km/h.
• Period 2: The leading vehicle would keep a constant speed of 40 km/h.
• Period 3: The leading vehicle would decelerate at 4 m/s2 until the speed was 30 km/h. • Period 4: The leading vehicle would keep a constant speed of 30 km/h.
• Period 5: The leading vehicle would accelerate at 2 m/s2 until the speed was up to 50 km/h.
• Period 6: The leading vehicle would keep a constant speed of 50 km/h.
• Period 7: The leading vehicle would decelerate at 2 m/s2 until the speed was 30 km/h. • Period 8: The leading vehicle would keep a constant speed of 30 km/h.
• Period 9: The leading vehicle would accelerate at 1.5 m/s2 until the speed was up to 40
km/h.
• Period 10: The leading vehicle would keep a constant speed of 40 km/h.
Figure 5: Car following scenario design
3.2.2.3 Lane-Changing Scenario
To investigate the differences between taxi drivers and non-professional drivers on lane-changing
behaviors, a road network was designed, as shown in Figure 6, which was also used in the previous
study (Yan and Wu, 2014). This road network was composed of a four-lane road segments with
the 80 km/h speed limit. Subjects would depart at the start point and get to the end point.
Additionally, two variable message signs (VMS) were installed in the network, which displayed
the current congestion state of the network. VMS-I showed that BEIJING ROAD was congested
(see Figure 7) and VMS-II showed that HANGZHOU ROAD was congested (see Figure 8). If
subjects wanted to choose the uncongested road, they needed to change from inside lane into
outside lane in the Test Segments. In order to avoid that other vehicles disturbed the experiment,
no other vehicle would appear in the Test Segments. Besides, each subject would drive three times
in the network and go through the Test Segments six times. Only when subjects had lane-changing
behaviors, the data was collected.
Figure 6: Lane-changing scenario design
Figure 7: VMS-I display
Figure 8: VMS-II display
3.2.3 High-Risk Scenario Design
At high-risk level, collision avoidance scenario was designed to examine driving behaviors when
facing an emergency crash, which was shown in Figure 9. The simulator vehicle entered the
signalized intersection during the green phase and drove normally. Meanwhile, the other system
vehicle was crossing this intersection from the right side at a velocity of 60 km/h during the red
phase. In order to create a right-angle crash at the intersection, a TTC sensor was used in this
scenario as well. TTC was set at seven seconds in advance of the TTC point. Therefore, when
subjects left 7 seconds temporal separation to the conflict point, TTC would trigger the system
vehicle to start to cross the intersection with a velocity of 60 km/h. According to the calculation,
the system vehicle was 116 meters away from the conflict point on the right side of the intersection.
Meanwhile, all other vehicles were cleared up before the simulator vehicle and the system vehicle
so that any other vehicle did not interfere with the drivers’ behavior. Thus, if subjects kept a present
speed along their presumed path to the conflict point, there would be a crash between the simulator
vehicle and the system vehicle. Additionally, all the roads were two-way with two lanes and the
speed limits were set at 80 km/h along the simulator path.
Figure 9: Collision avoidance scenario design
3.3 Experiment Procedure
Upon arrival, all subjects were asked to read and sign an informed consent form, which is approved
in China for the human subjects’ experiments. Each subject in this study was asked to take a short
training session, including the Traffic Regulation Education, the Safety Notice and the Familiarity
Training. In the Traffic Regulation Education session, each subject was advised to drive and
behave as they normally would and to follow traffic rules as they would in real-life situations. In
the Safety Notice session, each subject was told that they could quit the experiment at any time if
they had any motion sickness symptoms and any kind of discomfort. In the Familiarity Training
session, each subject was requested to have at least 10 minutes training to familiarize with the
driving simulator operation, including straight driving, acceleration, deceleration, left/right turns,
and other basic driving behaviors.
After completing the short training course, subjects could start formal experiments with two
scenarios in a random sequence so as to eliminate the time order effect. For security and liability
reasons, the escort would send every subject into the simulator cabin to commence the experiment
and keep an eye on the surveillance system during the whole experiment. Moreover, all subjects
were requested to rest at least 15 min between each scenario.
3.4 Subjects
First, researchers enlisted a group of about 55 people in this experiment, who were local drivers in
Beijing, China, with at least one year of driving experience. All taxi drivers were full-time job
drivers from different taxi companies. The other non-professional drivers were from different
vocations, such as teachers, company employees, students. Most subjects could finish the
experiment successfully but some subjects could not complete some parts of scenarios because of
the motion sickness so that they cannot follow the instructions. The gender and occupation
distribution for each scenario was shown in Table 1, which showed that the subject gender
distribution was balanced in each occupation group for each scenario.
Table 1: Gender and occupation distribution of each scenario
Scenario
Male Female
Taxi Non-professional Taxi Non-professional
Low-risk scenario 13 13 8 12 Medium-risk scenario Yellow indication Judgment 14 13 9 13 Car following 13 11 13 10 Lane-changing 16 14 9 13 High-risk scenario 14 13 9 13 25
Besides, the average age of all the subjects was 34, ranging from 20 to 52 years old (S.D. = 10).
Taxi drivers had an average mileage of 72.6 thousand kilometers per year and an average driving
experience of 15 years. By contrast, non-professional drivers had an average mileage of 12.6
thousand kilometers per year and an average driving experience of 4 years. Each subject was
required to finish all the scenarios in the driving simulator in order to get 500 Chinese RMB
(around 80 U.S. dollars) as a compensation for their participation.
CHAPTER FOUR: DATA COLLECTION
4.1 Simulator Data Collection Procedure
The driving simulator data collection included the experiment sampling time, vehicle speed,
acceleration, vehicle position, steering angle and many other related parameters. The data sampling
frequency is up to 60 Hz, and the collected raw data was stored in PLT type file, which was not
directly used for data analyses. First of all, PLT type files were manually converted to EXCEL
type files. Then, by using .NET framework, which is software developed by Microsoft that runs
primarily on Microsoft Windows, a program written by C Sharp language was developed to
automatically extract the experiment data, organize it, and easily process the raw data. From the
raw data, some significant variables were derived.
4.2 Low-Risk Scenario Data Collection
In the low-risk scenario design, there were three sub-scenarios, which were uphill and downhill
scenario (15 ° slope), right angle turn scenario, and S-type continuous curve scenario. Data
collection was based on 46 subjects’ behaviors in each sub-scenario (See Appendix A).
4.2.1 Uphill and Downhill Scenario Data Collection
The dependent measures for data analyses during the uphill period were defined as follows: 27
• USS (km/h): Uphill starting speed. The vehicle’s operating speed when the driver started to go uphill.
• UES (km/h): Uphill ending speed. The vehicle’s operating speed when the driver finished the uphill.
• USD (km/h): Uphill speed difference. The difference between the uphill starting speed and uphill ending speed, which was the uphill ending speed minus the uphill starting speed. • UAS (km/h): Uphill average speed. Average speed during the uphill period.
• UAD (m/s2): Uphill average deceleration. Average deceleration during the uphill period.
The dependent measures for data analyses during the downhill period were defined as follows: • DSP (km/h): Downhill starting speed. The vehicle’s operating speed when the driver
started to go downhill.
• DES (km/h): Downhill ending speed. The vehicle’s operating speed when the driver finished the downhill.
• DSD (km/h): Downhill speed difference. The difference between the downhill starting speed and downhill ending speed, which was measured as the downhill ending speed minus
the downhill starting speed.
• DAS (km/h): Downhill average speed. Average speed during the downhill period.
• DAA (m/s2): Downhill average acceleration. Average acceleration during the downhill
period.
4.2.2 Right Angle Turn Scenario Data Collection
The dependent measures for data analyses during right angle turn period were defined as follows: • RASS (km/h): Right angle starting speed. The vehicle’s operating speed when the driver
started to enter the right angle turn.
• RAES (km/h): Right angle ending speed. The vehicle’s operating speed when the driver finished the right angle turn.
• RASD (km/h): Right angle speed difference. The difference between the right angle starting speed and ending speed, which was measured as the right angle ending speed minus
the right angle starting speed.
• RAAD (m/s2): Right angle average deceleration. Average deceleration during the right
angle turn period.
• Out-of-lane (yes = 1; no = 0): Whether the vehicle went out of lane during the right angle turn period.
4.2.3 S-type Continuous Curve Scenario Data Collection
The dependent measures for data analyses during S-type continuous curve period were defined as
follows:
• SCSS (km/h): S-type continuous curve starting speed. The vehicle’s operating speed when the driver started to enter the S-type continuous curve.
• SCES (km/h): S-type continuous curve ending speed. The vehicle’s operating speed when the driver finished the S-type continuous curve.
• SCAS (km/h): S-type continuous curve average speed. Average speed during the S-type continuous curve period.
• SCDE (cm): type continuous curve average deviation. Average deviation during the S-type continuous curve period.
4.3 Medium-Risk Scenario Data Collection
4.3.1 Yellow Indication Judgment Scenario Data Collection
Data collection was based on each subject driving three test intersections in the road network. Each
subject would meet different test scenarios three times and the total of 49 subjects produced 147
records (See Appendix B). The related dependent measures for data analyses were defined as
follows:
• Speed (km/h): The vehicle’s operating speed at the onset of the yellow phase. • RLR (no = 0; yes = 1): Whether the driver ran a red light or not.
4.3.2 Car Following Scenario Data Collection
There were three kinds of periods in the car following process, including constant periods,
deceleration periods and acceleration periods (See Appendix C).
For the constant periods, there were a total of 5 periods. The dependent measures for data collection
in these periods were defined as follows:
• AS30 (km/h): Average speed during the constant period when the front vehicle is 30 km/h. • ASH30 (m): Average space headway between the simulator vehicle and the front vehicle
during the constant period when the front vehicle is 30 km/h.
• AS40 (km/h): Average speed during the constant period when the front vehicle is 40 km/h. • ASH40 (m): Average space headway between the simulator vehicle and the front vehicle
during the constant period when the front vehicle is 40 km/h.
• AS50 (km/h): Average speed during the constant period when the front vehicle is 50 km/h. • ASH50 (m): Average space headway between the simulator vehicle and the front vehicle
during the constant period when the front vehicle is 50 km/h.
For the deceleration periods, there were two periods during the car following process. The
dependent measures for data collection in each period were defined as follows:
• RT3 (s): Reaction time during the deceleration period 3, which was measured as the time between the start brake time of the front vehicle and the start brake time of the simulator
vehicle during the deceleration period 3.
• ADD3 (m/s2): Average deceleration during the deceleration period 3, which was measured
as the average deceleration during the deceleration period 3.
• RT7 (s): Reaction time during the deceleration period 7, which was measured as the time between the start brake time of the front vehicle and the start brake time of the simulator
vehicle during the deceleration period 7.
• ADD7 (m/s2): Average deceleration during the deceleration period 7, which was measured
as the average deceleration during the deceleration period 7.
For the acceleration periods, there were three periods during the car following process. The
dependent measures for data collection in each period were defined as follows:
• RT1 (s): Reaction time during the acceleration period 1, which was measured as the time between the start accelerate time of the front vehicle and the start accelerate time of the
simulator vehicle during the acceleration period 1. 32
• AAD1 (m/s2): Average acceleration during the acceleration period 1, which was measured
as the average acceleration during the acceleration period 1.
• RT5 (s): Reaction time during the acceleration period 5, which was measured as the time between the start accelerate time of the front vehicle and the start accelerate time of the
simulator vehicle during the acceleration period 5.
• AAD5 (m/s2): Average acceleration during the acceleration period 5, which was measured
as the average acceleration during the acceleration period 5.
• RT9 (s): Reaction time during the acceleration period 9, which was measured as the time between the start accelerate time of the front vehicle and the start accelerate time of the
simulator vehicle during the acceleration period 9.
• AAD9 (m/s2): Average acceleration during the acceleration period 9, which was measured
as the average acceleration during the acceleration period 9.
4.3.3 Lane-Changing Scenario Data Collection
Data collection in this scenario was based on each subject driving three times in the simulated road
network. Each subject was likely to change lanes in the Test segments. Finally, there were 136
lane-changing behaviors collected among all the subjects (See Appendix D). The related dependent
measures during the lane-changing period were defined as follows:
• LCT (s): Lane changing time, which was measured as lane-changing duration if a subject has a lane changing behavior.
• LCD (m): Lane changing distance, which was measured as the longitudinal distance of lane changing.
• LCS (km/h): Lane changing speed, which was measured as the average speed during lane changing period.
• MLS (m/s): Maximum lateral speed, which was measured as the largest lateral speed during lane changing period.
• ALS (m/s): Average lateral speed, which was measured as the average lateral speed during lane changing period.
• MLA (m/s2): Maximum lateral acceleration, which was measured as the largest lateral
acceleration during lane changing period.
• ALA (m/s2): Average lateral acceleration, which was measured as the average lateral
acceleration during lane changing period.
4.4 High-Risk Scenario Data Collection
In this scenario, data collection was based on 49 subjects’ behavior during the whole collision
period (See Appendix E). The dependent measures for data analyses were defined as follows:
• Speed (km/h): The vehicle’s operating speed at each second before the collision. Five seconds before the collision are recorded.
• Deviation (m): The vehicle’s lane deviation at each second before the collision, which was measured as distance between the vehicle location and the center of the lane. The positive
value means the vehicle is on the right side of the lane. Five seconds before the collision
are recorded.
• BD (m): The distance between the start brake location and the conflict point. The value could be 0 if drivers didn’t press the brake pedal.
• BS (km/h): The operating speed at the onset of the brake point. The value could be 0 if drivers didn’t press the brake pedal.
• BRT (s): Brake response time to collision, which was measured as the time between the start brake time and the expected collision time. The value could be 0 if drivers didn’t press
the brake pedal.
• DEC (m/s2): Average deceleration during the brake period. The value could be 0 if drivers
didn’t press the brake pedal.
• Crash (crash = 1; not = 0): Whether a crash occurred between the driver and the system vehicle.
CHAPTER FIVE: ANALYSES OF LOW-RISK SCENARIOS
This part of the thesis focuses on comparing driving behaviors between taxi drivers and
non-professional drivers in the low-risk scenarios. As explained before, the low-risk scenarios
consisted of three sub-scenarios: uphill and downhill analysis, right angle turn analysis, and S-type
continuous curve analysis. The hypothesis testing in the following analyses were based on a 0.1
significance level according to the limitations of the sample size.
5.1 Uphill and Downhill Analysis
For the uphill period, five dependent variables were chosen as potential factors, which might be
influenced by the subject occupation. The basic statistical descriptions of these five factors for taxi
drivers and nonprofessional drivers are listed in Table 2.
Table 2: Descriptive statistics of five factors related to the uphill period
Occupation Factor Mean Standard
Deviation Maximum Minimum
Taxi drivers USS 60.48 3.10 78.156 36.72 UES 38.56 2.89 65.844 25.92 USD 21.89 1.65 31.86 8.676 UAS 43.56 3.51 71.964 25.848 UAD -0.59 0.22 -0.13 -1.04 Non-professional drivers USS 66.71 2.70 86.148 46.512 UES 46.26 2.35 66.996 31.032 USD 20.45 1.53 30.636 6.84 UAS 52.52 3.08 74.124 32.724 UAD -0.64 0.23 -0.16 -1.06
An analysis of variance (ANOVA) for each factor is conducted to analyze whether there is a
difference between taxi drivers’ behaviors and non-professional drivers’ behaviors during the
uphill period (see Table 3).
Table 3: Analysis of variance (ANOVA) for dependent variables related to the uphill period
Variables Df Mean Square F-Value Sig.
USS 1 34.331 4.131 0.048 UES 1 51.926 7.614 0.008 USD 1 1.814 0.721 0.400 UAS 1 40.624 3.776 0.058 UAD 1 0.026 0.519 0.475 37
The ANOVA results show the significant difference between taxi drivers and non-professional
drivers on USS (p=0.048<0.1), UES (p=0.008<0.1), and UAS (p=0.058<0.1). The difference on
USS, UES, and UAS are shown in Figures 10, 11, and 12. For taxi drivers, the uphill starting speed
is only 60.48 km/h and the ending speed is only 38.56 as well. In comparison, both the uphill
starting speed and the ending speed of non-professional drivers are much higher than that of taxi
drivers. In addition, non-professional drivers also have a higher average speed during the uphill
period. However, there is no obvious difference in USD and UAD between taxi drivers and
non-professional drivers. The results indicate that taxi drivers had more driving experience and they
perform better in controlling vehicles so that they have lower starting speed and average speed.
Figure 10: Mean of the uphill starting speed (USS)
Figure 11: Mean of the uphill ending speed (UES)
Figure 12: Mean of uphill average speed (UAS)
Similarly, five dependent variables for the downhill period are also chosen to analyze the
difference between taxi drivers and non-professional drivers. The basic statistical descriptions for
the downhill period are listed in Table 4. Table 5 lists the ANOVA variance analysis for each
dependent variable, which shows that DSS, DES, and DAS are significant factors. Therefore, the 39
downhill scenario has the same conclusion with the uphill scenario, which shows taxi driver have
a better vehicle control than non-professional drivers.
Table 4: Descriptive statistics of five factors related to the downhill period
Occupation Factor Mean Standard
Deviation Maximum Minimum
Taxi drivers DSS 41.61 2.65 65.00 22.58 DES 63.92 2.34 76.43 42.22 DSD 22.30 2.29 32.04 0.62 DAS 50.94 2.00 64.69 34.98 DAA 2.39 0.25 3.73 0.06 Non-professional drivers DSS 52.03 2.11 70.86 39.53 DES 71.44 2.90 87.44 48.85 DSD 19.42 2.30 31.34 -3.24 DAS 59.43 2.72 77.98 37.92 DAA 2.53 0.33 4.39 -0.35
Table 5: Analysis of variance (ANOVA) for dependent variables related to the downhill period
Variables Df Mean Square F-Value Sig.
DSS 1 95.503 16.997 0.000 DES 1 49.885 7.055 0.011 DSD 1 7.342 1.398 0.243 DAS 1 63.528 10.838 0.002 DAA 1 0.017 0.203 0.655 40
5.2 Right Angle Turn Analysis
The basic statistical descriptions of the right angle turn for taxi drivers and nonprofessional drivers
are listed in Table 6. Table 7 lists the ANOVA variance analysis for each dependent variable,
which shows that there is no significant difference between taxi drivers and non-professional
drivers during the right angle turn.
Table 6: Descriptive statistics of five factors related to the right angle turn period
Occupation Factor Mean Standard
Deviation Maximum Minimum
Taxi drivers RASS 41.51 10.45 60.99 23.71 RAES 10.47 1.66 13.97 6.33 RASD 31.04 9.46 49.87 16.03 RAAD -0.12 0.23 0.23 -0.65 Non-professional drivers RASS 45.27 14.01 75.34 21.57 RAES 11.58 3.39 18.42 5.62 RASD 33.69 11.76 57.59 10.56 RAAD -0.13 0.34 0.48 -0.81
Table 7: Analysis of variance (ANOVA) for dependent variables related to the right angle turn period
Variables Df Mean Square F-Value Sig.
RASS 1 161.578 1.031 0.316
RAES 1 14.263 1.895 0.176
RASD 1 79.830 0.688 0.411
RAAD 1 0.001 0.012 0.913
The out-of-lane rate is also an important indication of the ability of vehicle control. 13 taxi drivers
out of 25 taxi drivers have an out-of-lane behavior during the right angle turn. However, 18
non-professional drivers out of 21 non-non-professional drivers drove out-of-lanes during the right angle
turn. Therefore, the probability of driving out of lanes for taxi drivers is:
p=13
25= 52%
and the probability of driving out of lanes for non-professional drivers is:
p=18
21= 85.7%
Z-test is applied to analyze whether there is a significant difference between two ratios. Figure 13
is the output from MINITAB for z-test with 95% confidence interval. The resulting p-value is
0.007, which is less than α=0.05. To conclude, there is significantly difference between taxi drivers
and non-professional drivers, which shows the out-of-lane behavior ratio of taxi drivers is smaller
than that of non-professional drivers.
Figure 13: MINITAB output: z-test for out-of-lane ratios
5.3 S-type Continuous Curve Analysis
Five dependent variables during the S-type continuous curve are measured to analyze a significant
difference between taxi drivers and non-professional drivers. The basic descriptive results are
descriptive in the Table 8.
Table 8: Descriptive statistical results of SCSS, SCES, SCAS, and SCDE for taxi drivers and non-professional drivers
Occupation Factor Mean Standard
Deviation Maximum Minimum
Taxi drivers SCSS 21.71 6.93 34.33 11.41 SCES 5.02 1.00 6.75 3.04 SCAS 5.33 .96 7.18 3.76 SCDE .67 .63 2.78 .15 Non-professional drivers SCSS 29.02 10.65 60.24 12.21 SCES 5.49 1.72 9.00 2.51 SCAS 5.90 1.56 8.95 3.57 SCDE .82 .56 2.30 .18
An analysis of variance (ANOVA) is also conducted of SPSS software to determine the statistical
significance of the four dependent variables. Table 9 lists the ANOVA results from SPSS, which
indicates that SCSS was a significant factor but SCES, SCAS, and SCDE are not under the 95%
confidence level. Therefore, taxi drivers have a smaller starting speed than non-professional
drivers (see Figure 14), which means taxi drivers have a better feeling of the potential risk when
they see a S-type curve in front. However, taxi drivers have the same ending speed, average speed,
and average deviation with non-professional drivers, indicating that taxi drivers have the same
performance on average speed and deviation with non-professional drivers in this condition.
Table 9: Analysis of variance (ANOVA) for dependent variables related to the S-type continuous curve period
Variables Df Mean Square F-Value Sig.
SCSS 1 609.43 7.277 0.010
SCES 1 2.566 1.230 0.273
SCAS 1 3.798 2.179 0.147
SCDE 1 0.040 1.611 0.211
Figure 14: Mean of the S-type continuous curve starting speed (SCSS)
CHAPTER SIX: ANALYSES OF MEDIUM-RISK SCENARIOS
This part of the thesis focuses on comparing driving behaviors between taxi drivers and
non-professional drivers in the medium-risk scenarios. As explained before, the medium-risk scenarios
consist of three sub-scenarios: yellow indication judgment analysis, car following analysis, and
lane-changing analysis. The hypothesis testing in the following analyses are based on a 0.1
significance level according to the limitations of the sample size.
6.1 Yellow Indication Judgment Analysis
In the Yellow Indication Judgment Scenario, two dependent variables are chosen as potential
factors, which might be influenced by the subject occupation. One is the speed at the onset of
yellow phase, which is a continuous variable and the other is the RLR, which is a categorical
variable.
6.1.1 Speed at the Onset of Yellow Phase Analysis
The basic statistical descriptions of operating speed for taxi drivers and non-professional drivers
at three test intersections are listed in Table 10. For all of taxi drivers, the mean speed at the onset
of yellow phase at all the three intersections is 69.46 km/h with a standard deviation of 15.37 km/h.
For all of non-professional drivers, the mean speed at the onset of yellow phase in all the three
intersections is 74.14 km/h with a standard deviation of 15.06 km/h. A two-sample t-test was
performed to compare the mean speed of the occupation at all three-test intersections. From the
results, the resulting p-value is 0.065 (t = 1.86), which is smaller than 0.1. Therefore, there is a
significant difference between taxi drivers and non-professional drivers in the operating speed. In
addition, a two-sample t-test was examined in each intersection respectively. It is also found that
the average operating speed for taxi drivers is significantly smaller than that of non-professional
drivers in each intersection (see Figure 15), which concurs with the study by Yan et al. (2014).
Table 10: Descriptive statistical results of speed at the onset of yellow phase for taxi drivers and non-professional drivers
Occupation TTI Mean
(km/h) N
Standard
Deviation Maximum Minimum
Taxi drivers 5s 70.36 23 14.67 115.38 40.02 5.5s 72.05 23 16.96 107.73 40.25 6s 65.97 23 14.37 97.22 43.94 Total 69.46 69 15.37 115.38 40.02 Non-professional drivers 5s 74.46 26 11.86 109.59 57.73 5.5s 78.25 26 18.77 150.87 42.38 6s 69.71 26 12.95 111.89 50.59 Total 74.14 78 15.06 150.87 42.38 47
Figure 15: Mean of the speed at the onset of the yellow phase
6.1.2 RLR Analysis
Red light running (RLR) is defined as an event that the vehicle is beyond the stop line at the onset
of red signal, which is one of the most important measures related to traffic safety. The basic
description of RLR for taxi drivers and non-professional drivers are summarized in Table 11. For
taxi drivers, 51 RLR violations were found at three intersections, which is 73.9% of all intersection
crossing behaviors. In comparison, only 25.6% of all intersection crossing behaviors among
non-professional drivers ran a red light. Using Chi-Square test to compare the RLR rate between taxi
drivers and non-professional drivers, it is found that the resulting p-value is 0.000 (𝜒𝜒1,1472 = 34.165 ), indicating that the RLR rate of taxi drivers is statistically larger than of non-professional drivers. Similarly, for each intersection, there is also a significant difference between taxi drivers
and non-professional drivers of RLR (for TTI = 5s: 𝜒𝜒1,492 = 5.728 , p = 0.017; for TTI =
5.5s: 𝜒𝜒1,492 = 14.750, p = 0.000; for TTI = 6s: 𝜒𝜒1,492 = 17.824, p = 0.000).
Table 11: Descriptive statistical results of RLR for taxi drivers and non-professional Drivers
Occupation TTI
RLR
Yes No
Count Row% Count Row%
Taxi drivers 5s 19 82.6% 4 17.4% 5.5s 17 73.9% 6 26.1% 6s 15 65.2% 8 34.8% Total 51 73.9% 18 26.1% Non-professional drivers 5s 13 50% 13 50% 5.5s 5 19.2% 21 80.8% 6s 2 7.7% 24 92.3% Total 20 25.6% 58 74.4%
Additionally, for taxi drivers, the RLR rate at 5 seconds TTI is larger than that at 5.5 seconds TTI,
which is also larger than that at 6 seconds TTI (82.6% Vs. 73.9% Vs. 65.2%). The downtrend also
happens on non-professional drivers. Therefore, as the TTI increases, the RLR rate shows a
downtrend on both taxi drivers and non-professional drivers. The reason is that as TTI increases,
the distance between the stop line and the simulator vehicle at the onset of yellow phase increases
so that drivers are more prone to stop before the stop line and have more time to decelerate. This
finding is consistent with some studies (Koll et al., 2004; Long et al., 2011; Yan et al., 2014) . 49
6.2 Car Following Analysis
In the car following scenario, three different periods are discussed separately, including constant
periods, acceleration periods, and deceleration periods. Each kind of period has its own
characteristics.
6.2.1 Constant period
In the constant period, there are two important variables reflect driving behaviors. One is average
speed during the constant period, the other one is average space headway during the constant
period. Table 12 shows the description of the average speed and the average space headway during
different constant periods.